Three-dimensional geological modeling methods and related devices based on neural networks and variation functions

Abstract

This invention belongs to the field of oil and gas field exploration and development, and discloses a three-dimensional geological modeling method and related apparatus based on neural networks and variograms. It constructs a structured three-dimensional geological modeling workflow, defining a complete sequence of steps from data input to model output. An initial model based on neural networks and variograms is trained using preprocessed multi-source geological data to generate a parameter generation model that automatically outputs key parameters for each stage of the workflow. When executing the workflow, the parameters are invoked, and the intermediate geological model is quantitatively evaluated. If the model does not meet the standards, the parameters are fine-tuned and the process is repeated until a satisfactory three-dimensional geological model is output. This method significantly improves modeling efficiency, avoids inconsistencies caused by expert subjective differences, enhances model repeatability and objectivity, reduces labor costs, and meets the needs of high efficiency and large-scale oil and gas field exploration and development.

Description

Technical Field

[0001] This invention belongs to the field of oil and gas field exploration and development technology, and in particular relates to a three-dimensional geological modeling method and related apparatus based on neural networks and variation functions. Background Technology

[0002] In the field of oil and gas field exploration and development, geological modeling is a core link connecting geological research and development practice. Its accuracy directly affects key tasks such as oil and gas reserve assessment, development plan design, and optimization of extraction efficiency. Currently, traditional geological modeling workflows generally include key steps such as structural modeling, facies modeling, and attribute modeling. Each step often employs stochastic modeling algorithms such as Kriging interpolation, sequential indicator simulation, and sequential Gaussian simulation to construct the geological model. These algorithms have been widely used in the industry and have become the mainstream technical means for geological modeling. However, the effectiveness of algorithm application, in addition to the characteristics of the algorithm itself, depends on the reasonable setting of variogram parameters (including nugget value, sill value, range, etc.). This parameter setting is a core and critical link in the traditional modeling workflow.

[0003] In existing traditional modeling workflows, the determination of variogram parameters mainly relies on manual adjustment, a process heavily dependent on the personal experience and subjective judgment of geological experts. Due to differences in the understanding and experience of different geological experts regarding geological bodies, the variogram parameter settings for the same geological body are often inconsistent. This not only hinders the standardization and efficiency improvement of geological modeling workflows but also results in discrepancies in model results constructed under the same geological conditions, severely impacting the repeatability and objectivity of the model results. Furthermore, manual parameter adjustment requires repeated iterative verification, which is time-consuming and labor-intensive, further exacerbating the inefficiency of the modeling workflow and failing to meet the actual needs of large-scale and efficient oil and gas field exploration and development.

[0004] It is evident that in the traditional oil and gas field geological modeling process, the variogram parameters of the stochastic modeling algorithm rely on manual debugging and expert experience, resulting in problems such as low modeling efficiency, poor consistency of results, and insufficient repeatability and objectivity. Summary of the Invention

[0005] This invention provides a three-dimensional geological modeling method and related apparatus based on neural networks and variograms to solve the problems of low modeling efficiency, poor consistency of results, and insufficient repeatability and objectivity caused by the variogram parameters of stochastic modeling algorithms relying on manual debugging and expert experience.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A three-dimensional geological modeling method based on neural networks and variograms includes: Based on the geological background of the work area and the modeling objectives, a structured 3D geological modeling workflow is constructed; the 3D geological modeling workflow is used to reflect the complete sequence of steps and dependencies from data input to 3D geological model output; The pre-constructed initial network model is trained using pre-processed multi-source geological data to obtain a trained parameter generation model. Based on the parameter generation model, key parameters required for each stage of the structured 3D geological modeling workflow are generated. The initial network model is constructed based on a neural network and a variation function. The process involves calling key parameters and executing a structured 3D geological modeling workflow. During the execution of the structured 3D geological modeling workflow, a quantitative quality assessment is performed on the generated intermediate geological model. If the assessment indicators do not meet the preset standards, a parameter generation model is used to fine-tune the parameters. The structured 3D geological modeling workflow and assessment process are repeated until a 3D geological model that meets the quality requirements is generated, and finally, the 3D geological model is output.

[0007] Furthermore, in the step of constructing a structured 3D geological modeling workflow based on the geological background and modeling objectives of the work area, the structured 3D geological modeling workflow includes a workflow parsing process, an execution sequence arrangement process, and a task scheduling and execution process; wherein: The workflow parsing process includes: reading the workflow steps preset by the graphical interface or script, converting the complete process of data loading → surface modeling → phase modeling → attribute modeling → model verification and output into a logical structure that the system can recognize and process, and marking the task nodes and their sequential dependencies; The execution order arrangement process includes: planning the optimal execution route based on task dependencies, and arranging parallel execution for tasks with no dependencies; The task scheduling and execution process includes: automatically starting each modeling stage according to the planned sequence, passing the output results of the previous task to the next task, and monitoring the running, completion, and failure status of each task in real time to achieve automated workflow.

[0008] Furthermore, in the step of training the pre-constructed initial network model using pre-processed multi-source geological data, the multi-source geological data includes well location data, well trajectory data, well logging curve data, well stratification data, seismic interpretation stratigraphic data, and seismic inversion data volumes; the preprocessing of the multi-source geological data includes data type standardization, spatial alignment and datum unification, quality control, well logging curve coarsening, standardization, and data fusion; specifically as follows: The fields of the multi-source geological data are supplemented and the units are labeled according to the preset specifications, and the outliers are filled with standard identifiers. All data are unified to the target projection coordinate system of the work area, and the logging depth is converted into absolute vertical depth through well trajectory data; Invalid data and extreme points were removed through integrity checks, outlier filtering using the interquartile range method, and statistical analysis. For continuous attribute data, the arithmetic mean method is used, and for discrete attributes, the mode method is used to map well logging attributes to three-dimensional grid cells. Perform Zscore standardization on various types of data; Standardized data is spliced ​​together along the channel dimension to construct a four-dimensional input tensor, thereby completing the data fusion.

[0009] Furthermore, the step of training the pre-constructed initial network model using pre-processed multi-source geological data includes: The pre-collected historical work area sample set is preprocessed into multi-source geological data to serve as training data. The output label adopts the optimal parameter combination of the variogram function, which includes the primary, secondary, and vertical range, the primary and secondary azimuth angles and vertical dip angle, the sill value, and the nugget value. A supervised learning approach is adopted, using backpropagation algorithm and gradient descent optimizer, with mean squared error as the loss function, to minimize the difference between the predicted value and the true label, so that the initial network model learns the nonlinear mapping relationship between data features and variogram parameters, and outputs parameters to generate the model; After training, through forward propagation, feature extraction from convolutional layers, introduction of nonlinearity using the ReLU activation function, dimensionality reduction using pooling layers, and mapping using fully connected layers, the predicted values ​​of the variogram parameters are output as key parameters required for each stage of the structured 3D geological modeling workflow.

[0010] Furthermore, the construction modeling step of the structured 3D geological modeling workflow employs the Kriging interpolation algorithm, with the specific steps as follows: A spherical variogram model is constructed using the predicted values ​​of the variogram parameters output by the parameter generation model; the predicted values ​​of the variogram parameters include the sill value, nugget value, and range parameter. By constructing a spatial coordinate transformation matrix using azimuth and dip angles, anisotropic space is converted into isotropic space. The weights are obtained by solving the Kriging equations. The stratigraphic data are then smoothed and coarsely interpolated to generate a structural surface, thus completing the structural modeling process.

[0011] Furthermore, the phase modeling step of the structured 3D geological modeling workflow employs a sequential instruction simulation algorithm, with the specific steps as follows: Define indicator functions to distinguish different lithofacies or sedimentary facies types, and use a parameter generation model to generate range and sedimentary facies zone extension direction parameters for each facies or facies combination; Traverse all grid nodes to be simulated, use the indicator kriging algorithm, call the variogram parameters generated by the parameter generation model to construct the covariance matrix, and calculate the conditional probability that the current node belongs to the corresponding phase type. The phase type of a node is determined from the cumulative distribution function by Monte Carlo sampling, and the node is used as a known point in the calculation of the next node to generate a three-dimensional phase model.

[0012] Furthermore, in the step of quantitatively evaluating the quality of the generated intermediate geological model, the quality evaluation indicators of the quantitative quality evaluation include the matching error between the structural layer and the well point stratification elevation, the consistency between the formation thickness and the known geological statistical thickness, the similarity between the attribute model and the well logging curve histogram, and the matching accuracy package between the simulated value and the actual well logging value at the well location. If any indicator exceeds the preset threshold, a feedback loop is automatically triggered, the parameter generation model is called again to fine-tune the modeling parameters, a new parameter set is generated to drive a new round of modeling calculations, and the iteration continues until all quality assessment indicators meet the requirements.

[0013] A three-dimensional geological modeling system based on neural networks and variograms includes: The workflow construction module is used to construct a structured 3D geological modeling workflow based on the geological background of the work area and the modeling objectives; the 3D geological modeling workflow is used to reflect the complete sequence of steps and dependencies from data input to 3D geological model output; The model training module is used to train a pre-constructed initial network model using pre-processed multi-source geological data to obtain a trained parameter generation model, and to generate key parameters required for each stage of the structured 3D geological modeling workflow based on the parameter generation model; wherein, the initial network model is constructed based on a neural network and a variation function; The execution module is used to call key parameters and execute the structured 3D geological modeling workflow. During the execution of the structured 3D geological modeling workflow, the generated intermediate geological model is quantitatively evaluated. If the evaluation indicators do not meet the preset standards, the parameter generation model is used to fine-tune the parameters. The structured 3D geological modeling workflow and evaluation process are repeated until a 3D geological model that meets the quality requirements is generated, and finally the 3D geological model is output.

[0014] A three-dimensional geological modeling device based on neural networks and variograms, comprising: Memory, used to store computer programs; A processor is used to implement the steps of the above-described three-dimensional geological modeling method based on neural networks and variation functions when executing the computer program.

[0015] A computer-readable storage medium storing a computer program, which, when executed by a processor, is used to implement the steps of the above-described three-dimensional geological modeling method based on neural networks and variograms.

[0016] Compared with the prior art, the present invention has the following beneficial effects: This invention provides a 3D geological modeling method based on neural networks and variograms. It constructs a structured 3D geological modeling workflow, defining a complete sequence of steps from data input to model output. An initial model based on neural networks and variograms is trained using preprocessed multi-source geological data, generating a parameter generation model to automatically output key parameters for each stage of the workflow. During workflow execution, parameters are invoked, and the intermediate geological model undergoes quantitative quality assessment. If the model fails to meet standards, parameters are fine-tuned, and the process is repeated until a satisfactory 3D geological model is output. The neural network automatically derives variogram parameters by learning geological data patterns, reducing manual intervention. The variograms ensure accurate representation of geological spatial variability, and the structured workflow standardizes the process. The evaluation and fine-tuning mechanism improves model accuracy through iterative optimization. This method significantly improves modeling efficiency, avoids inconsistencies caused by expert subjective differences, enhances model repeatability and objectivity, reduces labor costs, and meets the demands for high efficiency and large-scale oil and gas field exploration and development. Attached Figure Description

[0017] Figure 1 A schematic diagram illustrating the implementation process of a three-dimensional geological modeling method based on neural networks and variograms provided in this embodiment of the invention; Figure 2 The three-dimensional geological model provided in this embodiment of the invention is constructed using a three-dimensional geological modeling method based on neural networks and variograms. Figure 3 A flowchart illustrating a three-dimensional geological modeling method based on neural networks and variograms, provided for an embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of a three-dimensional geological modeling system based on neural networks and variograms, provided for an embodiment of the present invention. Detailed Implementation

[0018] To further understand the content of this invention, the invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments are merely illustrative and not limiting of the invention.

[0019] As mentioned in the background section, in the traditional modeling process, the effectiveness of stochastic modeling algorithms (such as Kriging interpolation, sequential indicator simulation, and sequential Gaussian simulation) used in key steps such as structural modeling, phase modeling, and attribute modeling depends not only on the algorithm itself, but also on the manual adjustment of parameters such as variogram parameters (null value, sill value, range, etc.). This process relies heavily on the personal experience of geological experts, which restricts the standardization and efficiency of modeling, and also affects the repeatability and objectivity of model results.

[0020] To achieve the aforementioned objectives, this embodiment provides a 3D geological modeling method based on neural networks and variograms. This method, based on existing standardized modeling workflows, focuses intelligent computing technology combining variograms and neural networks on determining algorithm parameters for each core step, achieving a key shift from "manual parameter setting" to "system-intelligent parameter generation." This method effectively solves the problems of low efficiency and poor consistency caused by reliance on expert experience in traditional processes, significantly improving automation while maintaining high process transparency and geological rationality. Its standard workflow-based design makes it easy to integrate into existing platforms, and the intelligent parameter generation enhances the controllability and operability of the modeling process.

[0021] like Figure 3 As shown, this embodiment provides a three-dimensional geological modeling method based on neural networks and variograms, including: Based on the geological background of the work area and the modeling objectives, a structured 3D geological modeling workflow is constructed; the 3D geological modeling workflow is used to reflect the complete sequence of steps and dependencies from data input to 3D geological model output; The pre-constructed initial network model is trained using pre-processed multi-source geological data to obtain a trained parameter generation model. Based on the parameter generation model, key parameters required for each stage of the structured 3D geological modeling workflow are generated. The initial network model is constructed based on a neural network and a variation function. The process involves calling key parameters and executing a structured 3D geological modeling workflow. During the execution of the structured 3D geological modeling workflow, a quantitative quality assessment is performed on the generated intermediate geological model. If the assessment indicators do not meet the preset standards, a parameter generation model is used to fine-tune the parameters. The structured 3D geological modeling workflow and assessment process are repeated until a 3D geological model that meets the quality requirements is generated, and finally, the 3D geological model is output.

[0022] The prediction method provided in this embodiment will be further explained below with reference to the accompanying drawings: like Figure 1 As shown, this embodiment provides a 3D geological modeling method based on neural networks and variograms. Specifically, it proposes an automated 3D geological modeling method that relies on a preset workflow and introduces a neural network to intelligently calculate variogram parameters. The core idea is as follows: when setting the modeling process, the user still follows the conventional workflow sequence (e.g., data preparation → structural modeling → facies modeling → attribute modeling → model output). However, during the process, except for basic parameters such as grid step size which need to be manually set, other key modeling parameters (such as variogram parameters, stochastic simulation path control parameters, etc.) no longer rely on manual experience. Instead, they are automatically generated by the integrated neural network module through intelligent calculation, thereby driving the modeling engine to automatically construct the 3D geological model. The specific steps of this method are as follows: Step 1: Parsing, planning, and execution engine based on preset workflows: Users pre-define a structured modeling workflow based on the geological background and modeling objectives of the work area, using a graphical interface or scripts. This workflow clarifies the complete sequence of steps from data input to model output and their dependencies (e.g., data loading -> structural surface modeling -> facies modeling -> attribute modeling -> model verification and output).

[0023] Step 2: Intelligent calculation of key parameters based on neural mesh and variogram: The parameter generation model, the core innovation of this system, aims to transform the traditional parameter setting process, which relies on manual experience, into a data-driven intelligent decision-making and automatic generation process. This module introduces a neural network-driven variogram parameter calculator to automatically calculate and optimize key parameters in each modeling stage of the preset workflow, significantly improving modeling efficiency and reliability.

[0024] In the intelligent calculation of structural modeling parameters, it is usually necessary to generate three-dimensional structural layers within a specified boundary using Kriging interpolation based on well point structural layer data and seismic interpretation horizons. The setting of variogram parameters (such as range and nugget value) directly affects the results. This method calls the calculation interface through a workflow engine, inputting standardized seismic and well data into a pre-trained convolutional neural network (CNN) model. This CNN model automatically predicts the optimal combination of variogram parameters by analyzing the spatial structural features in the data and returns them to the modeling process for layer interpolation, effectively reducing the subjectivity of manual settings.

[0025] In the intelligent calculation of facies modeling parameters, algorithms such as sequential indicator simulation rely heavily on the settings of variogram parameters for each facies and the global facies scale model to restore the three-dimensional spatial distribution of different lithofacies or sedimentary facies within the reservoir. This method triggers the corresponding interface through a workflow engine, taking wellpoint facies calibration data, seismic attribute volumes, and sedimentary pattern constraints as input, and passing them to a pre-trained convolutional neural network (CNN) model. This CNN automatically outputs the optimal variogram parameters and facies probability weights for each facies by analyzing the complex relationship between the spatial structural features of the input data and the facies distribution pattern, thereby generating a three-dimensional facies model with clear geological significance and reasonable facies band continuity.

[0026] In the intelligent calculation of attribute modeling parameters, after the well logging curves are coarsened, spatial interpolation is performed using stochastic algorithms such as sequential indication and sequential Gaussian simulation. Many of these control parameters traditionally rely on expert experience for setting. This method triggers the attribute modeling parameter calculation interface through a workflow engine, inputting the coarsened attribute data and geological constraints into a pre-trained convolutional neural network (CNN) model. This CNN model learns the statistical characteristics and spatial distribution patterns of known data from the work area, automatically recommending simulation parameters that conform to geological laws. This generates an attribute model that is both faithful to the well data and geologically reasonable, improving model accuracy and automation.

[0027] Step 3: Automated workflow execution and closed-loop iterative optimization: The system integrates a pre-defined workflow with a parameter generation model to automate and optimize 3D geological modeling. It combines the task sequence defined in the workflow with a parameter set generated by an intelligent algorithm, driving the underlying modeling engine to execute each modeling stage sequentially by calling the modeling software's application programming interface. During this process, the integrated model quality evaluation unit automatically performs a quantitative evaluation of the newly generated model, calculating key indicators such as its matching error with known data (e.g., well point data) and its conformity to geological statistics. The system incorporates a feedback loop mechanism to form a closed-loop optimization. When evaluation indicators (e.g., errors exceeding thresholds or insufficient geological rationality) fail to meet preset standards, the system automatically triggers parameter adjustment commands, feeding back to the parameter generation model for fine-tuning or model retraining, and initiating a new round of modeling. This cycle ultimately constructs a complete automated "modeling-evaluation-optimization" closed loop, effectively improving model accuracy and efficiency. For example, key steps in Step 1 include workflow parsing: During workflow parsing, the system reads and understands the user-preset workflow steps, transforming the entire process (e.g., data loading → surface modeling → phase modeling → attribute modeling) into a logical structure that it can recognize and process internally, indicating task nodes and their dependencies. Execution order arrangement: After understanding the overall workflow, the system intelligently plans the optimal execution route for tasks, ensuring that a task is only started after all its prerequisite tasks are completed; for tasks without dependencies, they are scheduled to run simultaneously to improve overall efficiency. Task scheduling and execution: Each modeling stage is automatically and sequentially started according to the planned order. It is responsible for accurately transmitting the output of the previous task to the next task and monitoring the execution status of each task in real time (e.g., running, completed, failed). The entire process resembles an automated assembly line; when one stage successfully completes, the system automatically starts the next stage until the entire workflow is complete.

[0028] For example, in step two, the powerful spatial feature learning capabilities of CNNs are utilized to intelligently predict key algorithm parameters in structural modeling, facies modeling, and attribute modeling. Training samples are constructed by systematically collecting multi-source geological data (including well data, seismic data, and geological models) from the work area, along with their corresponding optimal parameter sets obtained through expert verification or traditional methods. After preprocessing such as data alignment and standardization, the data is fused into a four-dimensional tensor, which serves as input to a dedicated CNN model to learn the complex nonlinear mapping from complex geological data to optimal parameters. The designed CNN model is tailored to the characteristics of three-dimensional data. Its architecture includes convolutional layers that extract spatial features using three-dimensional convolutional kernels, activation layers that introduce nonlinearity, pooling layers that perform downsampling, and fully connected layers that output predicted parameters. The model is trained using a backpropagation algorithm, with mean squared error as the loss function, allowing the predicted parameters to continuously approach the true optimal parameter labels. After training, when faced with new work area data, the same preprocessing is applied before inputting it into the model, directly outputting a recommended set of key parameters, which then drives traditional algorithms such as Kriging interpolation or sequential indication to automatically generate geological models.

[0029] For example, the calculation of intelligent modeling parameters in step two is highly dependent on the quality of data preprocessing. This process aims to clean, correct, and fuse heterogeneous data (wellbore, seismic, geological stratification) from multiple sources and at multiple scales to construct a standard dataset with unified spatial coordinates and reliable quality, in order to meet the application requirements of deep learning model training and geostatistical algorithms.

[0030] In this embodiment, the preprocessed dataset mainly includes the following six types of basic data: (1) Well location data: Data description: Basic point data defining the wellhead's geographical location, elevation datum, and well body structural attributes.

[0031] Data specifications: Well name: A unique identifier for the well shaft, serving as the primary key for linking other data.

[0032] Wellhead coordinates: Planar projection coordinates of the wellhead in the work area coordinate system (unit: m).

[0033] Core filling elevation: The elevation (in meters) of the wellhead core filling relative to the reference surface (usually sea level), used for depth-elevation correction.

[0034] Well type: The engineering or geological classification of a well (e.g., production well, appraisal well, exploratory well, etc.).

[0035] Bottom-of-well coordinates: The projected coordinates of the target point or endpoint on the horizontal plane.

[0036] Complete drilling depth: The final measured depth of the wellbore (unit: m).

[0037] (2) Well trajectory data description: Continuous measurement data describing the geometric shape of the wellbore in three-dimensional space.

[0038] Data specifications: Depth measurement: The cumulative length along the wellbore axis (unit: m).

[0039] Well inclination angle: The angle between the tangent to the wellbore axis and the vertical line (unit: °).

[0040] Azimuth: The angle between the projection of the wellbore axis onto the horizontal plane and the due north direction (unit: °).

[0041] Dogleg degree: The rate of change of curvature of the wellbore trajectory within a unit well section (unit: ° / 30m or ° / m), used to assess the degree of wellbore curvature.

[0042] (3) Well logging curve data Data Description: The formation geophysical response data continuously collected along the wellbore trajectory is the core basis for lithology identification and parameter inversion.

[0043] Data specifications: Sampling depth (Depth): The measurement depth along the wellbore trajectory, usually using equal-interval sampling (unit: m).

[0044] Attributes include, but are not limited to, natural gamma (GR), deep / shallow lateral resistivity (RT), volume density (DEN), and acoustic transit time (AC).

[0045] Outlier handling: Missing or invalid measurements should be filled with standard identifiers (such as -999.999).

[0046] (4) Well stratification data: Data description: Structured data identifying the depth of geological stratigraphic interfaces within the wellbore.

[0047] Data specifications: Layer name: The standard name of a geological stratum or stratigraphic interface.

[0048] Depth: The depth value of the top or bottom boundary of the stratum (unit: m).

[0049] Well Name: Contains a well name field, used to construct a stratigraphic correlation framework between multiple wells.

[0050] (5) Seismic interpretation stratigraphic data: Data description: Discretized surface data (usually regular or irregular grids) reflecting the spatial morphology of underground geological interfaces.

[0051] Data specifications: Grid coordinates: X and Y coordinates of planar grid nodes (unit: m).

[0052] Elevation: The vertical domain value (elevation depth) of the stratigraphic interface at the corresponding grid node. The elevation depth downwards is either negative or positive, requiring a unified work area definition.

[0053] (6) Seismic inversion data volume: Data Description: Three-dimensional volume data generated by seismic inversion algorithms, reflecting the physical properties of underground rocks (such as wave impedance, porosity, etc.).

[0054] Data specifications: 3D Mesh: Includes a 3D index of Inline (line number), Crossline (track number), and Time / Depth (time / depth).

[0055] Volume attribute value: The physical attribute value corresponding to the mesh cell.

[0056] In this embodiment, data preprocessing is the foundation for building intelligent models. The heterogeneous raw data is transformed into standardized feature tensors through the following five steps.

[0057] (1) Spatial alignment and datum plane unification: Core objective: To ensure that all data (well data, seismic interpretation horizon data, seismic inversion data volumes) are located in a unified coordinate system and under a vertical reference plane.

[0058] Planar coordinate system one: The well location coordinates, seismic interpretation horizons, and seismic inversion volumes are uniformly converted to the target projection coordinate system of the work area (such as UTM or Beijing 54 coordinate system).

[0059] Vertical reference plane unification: The measured depth (MD) is calculated using well trajectory data (azimuth and dip) and converted into the absolute vertical depth (TVDSS) relative to mean sea level (MSL).

[0060] Ensure that the depth system of the seismic inversion body and the interpretation horizon is consistent with the well logging TVDSS to eliminate horizon misalignment caused by differences in core elevation (KB).

[0061] (2) Data quality control: Before the data is entered into the model, outliers are removed using statistical methods: Completeness check: Verify whether the well location data is missing (such as missing core height, missing well trajectory, etc.).

[0062] Outlier filtering: Identify anomalous noise values ​​in well logging and seismic attributes using the interquartile range (IQR) method.

[0063] Statistical analysis: Draw histograms and scatter plots. If the distribution of well logging attributes shows abnormal extreme points, truncation is required; check if there are large areas of invalid values ​​in the seismic inversion body.

[0064] (3) Well logging curve coarsening: Since the well logging sampling rate is much higher than the resolution of the geological grid, it is necessary to map the well logging attributes on the well trajectory to the three-dimensional grid cells: Continuous properties (such as porosity): Arithmetic mean method is used. Calculation formula:

[0065] in, This represents the number of well logging sampling points that fall within the grid.

[0066] Discrete properties (such as lithofacies): The mode method is used to take the facies state with the highest frequency in the grid as the value of the grid to ensure the representativeness of the lithological framework.

[0067] (4) Standardization: Provides stable input for CNN prediction: The core objective is to perform Z-score standardization on the elevation values ​​(Z) of seismic interpretation horizons, well-layer elevation values ​​(Z), well logging data (such as GR, AC, DEN, etc.), and attribute values ​​of seismic inversion data volumes. This is primarily to ensure that the data input to the CNN model has a consistent dimension and distribution, eliminating data differences from different sources, with different dimensions and numerical ranges, thereby improving the numerical stability and efficiency of the model's predictions, rather than directly using it for CNN model training.

[0068] Calculation formula:

[0069] In the formula, It is the standardized value; It is the original value; The mean of the dataset; denoted as the standard deviation of the dataset.

[0070] Importance: Standardized data helps trained CNN models extract spatial features more stably and accurately, thereby outputting more reliable variogram parameters.

[0071] (5) Data fusion: Constructing the feature input tensor of the CNN: Core objective: To construct a four-dimensional input tensor acceptable to CNN models by stitching together standardized and aligned seismic interpretation stratigraphic data, seismic inversion attribute volumes (reflecting macroscopic structural morphology), well layers, and well logging curve values ​​(providing local precise control) in the channel dimension.

[0072] Tensor structure:

[0073] : Represents width × height, which is the dimension of a planar grid; Depth, vertical spatial dimension (D=1 for single-layer analysis, and can be stacked along this dimension for multi-layer systems). : Number of feature channels (corresponding to different geological / geophysical properties).

[0074] Function: The fused tensor fully represents the spatial structure characteristics and attribute distribution patterns of the work area, and serves as the direct input carrier for the CNN model to predict variogram parameters and extract geological features.

[0075] For example, the core of calculating the intelligent modeling parameters in step two lies in: using a pre-trained CNN model, based on the pre-processed fused data, to intelligently predict the variogram parameters required for Kriging interpolation, sequential indicator simulation, and sequential Gaussian simulation. The CNN model pre-training includes: 1. Training Data Preparation: Construct a large-scale historical work area sample set, where each sample consists of input features (X) and output labels (Y): Input Features (X): The preprocessed seismic inversion volume, the coarsened well logging attribute volume, and the stratigraphic feature field are fused through multiple channels to form a four-dimensional tensor. This represents the specific spatial heterogeneity of the work area.

[0076] Output label (Y): The optimal parameter combination corresponding to the work area, verified by experts or repeatedly optimized through traditional methods (such as experimental variation function fitting and step size adjustment). ,in Primary, secondary, and vertical range shifting; : Principal direction azimuth and vertical tilt; : Sill value; : Value of a nugget.

[0077] 2. Model Training and Learning: Supervised learning is employed to train the CNN model using the aforementioned dataset. Essentially, this involves teaching the model the complex nonlinear mapping relationship between input data features and the optimal variogram parameters. Through backpropagation and an optimizer, the network's internal parameters are continuously adjusted to minimize the difference between the model's predicted values ​​and the true labels (e.g., the mean squared error loss function), until the model can accurately capture the correlation between the spatial structure features in the data and the variogram parameters.

[0078] Explainable, a complete CNN prediction of variogram parameters primarily involves forward propagation (used for final parameter prediction and also throughout the inference phase of model training) and backpropagation (specifically for model training to optimize network weights). The goal of backpropagation is to optimize parameters such as weights (W) and biases (b) within the network, improving the CNN's prediction output. As close as possible to the optimal parameter labels verified by geological experts or determined by traditional methods. Backpropagation includes calculating the loss function, updating weights, and backpropagation.

[0079] Loss function: The loss function quantifies the difference between the model's predicted value and the true label, using mean squared error for quantification.

[0080] The calculation formula is:

[0081] In the formula, Loss value; The number of training samples in a batch; ; : Label of the true variation function parameter of the i-th sample.

[0082] The goal of training is to minimize the loss value L by adjusting the model parameters until the model can accurately capture the spatial correlation distance (range) and extension direction (azimuth) of geological entities from seismic textures and well point distribution.

[0083] Weight Update and Backpropagation: To minimize the loss function, the network parameters need to be updated using the gradient descent algorithm. The gradient (partial derivative) of the loss function with respect to the weights (W) and biases (b) of each layer is calculated. The backpropagation algorithm efficiently computes these gradients using the chain rule.

[0084] The calculation formula is:

[0085] In the formula, The learning rate controls the step size of each parameter update. : Loss function relative to the first The gradient of the layer weights. This gradient indicates the direction and magnitude in which the weights should be adjusted to reduce the loss.

[0086] 3. Once the model is trained, only forward propagation is needed. The new work area data is input into the trained CNN model to predict the variogram parameters. The predicted values ​​can then be directly used in traditional interpolation algorithms for calculation. Forward propagation is fundamental to the model's prediction process. Preprocessed geological data tensors are input into the network, and the data sequentially passes through convolutional layers, activation functions, pooling layers, and fully connected layers, ultimately outputting the predicted values ​​of the variogram parameters.

[0087] Convolutional layer computation: Convolutional layers are the core of feature extraction. Data flows through multiple 3×3×3 convolutional layers, extracting spatially relevant information such as layer undulations, seismic impedance changes, and well point attributes layer by layer.

[0088] The calculation formula is:

[0089] In the formula, : is the first Output feature maps of convolutional layers; : No. The weight matrix of the convolution kernel (filter) of the layer. : No. The output of layer -1 is also the first layer. The input to the layer. The first layer is the fused data tensor formed after preprocessing. : No. Layer bias terms.

[0090] Activation function: The ReLU activation function introduces non-linearity, enabling it to learn complex patterns.

[0091] The calculation formula is:

[0092] In the formula, The th after activation function The final output of the layer; : The linear computation result of the current layer (i.e., the output of the convolutional layer); the ReLU function is effective in alleviating the gradient vanishing problem and accelerating convergence.

[0093] Pooling layer calculation: Pooling layers are used for dimensionality reduction and preservation of key anisotropic features.

[0094] The formula for calculating max pooling is:

[0095] In the formula : is the output feature map after pooling at the position The value; : Corresponding to the input feature map A local area on the surface. : Input feature map at location The value of .

[0096] Max pooling extracts the maximum value within a local region, thus reducing the amount of data while preserving the most salient features.

[0097] Fully connected layer and output: After multiple rounds of convolution and pooling, the data is flattened and fed into the fully connected layer, which finally outputs the predicted values ​​of the variation function parameters.

[0098] The calculation formula is:

[0099] In the formula, The model's predicted output vector typically contains three components, corresponding to the key parameters of the strain function. For example, for a spherical model, these are the range, nugget value, and abutment value. Weights and biases of the output layer (i.e., the Lth layer). : The output feature of the last hidden layer (the flattened vector).

[0100] For example, in step two, the construction modeling based on key parameters of neural mesh and variogram uses Kriging interpolation, and the construction modeling mainly focuses on smoothing and accurately interpolating the layer data.

[0101] Algorithm principle: Using ordinary kriging, its estimated value for: in, The weight to be determined.

[0102] Parameter application: Range of CNN output and direction Used to construct variation function models (such as spherical models): , where the distance h is the spatial distance after the anisotropic rotation matrix transformation.

[0103] According to CNN's predictions, Construct a spatial coordinate transformation matrix to convert anisotropic space into isotropic space.

[0104] Solving the Kriging equations yields This generates high-precision constructed surfaces.

[0105] For example, step two, the key parameter intelligent facies modeling based on neural meshes and variograms, is implemented using sequential indicator simulation. Facies modeling handles discrete variables (such as lithofacies 0, 1, 2). For K lithofacies, an indicator function is defined:

[0106] CNNs calculate specific range and orientation for each phase (or combination of phases).

[0107] main direction : Represents the direction of extension of the sedimentary facies zone; Variable range : Represents the average size of the phase.

[0108] Simulation steps: Define random pathfinding: traverse all nodes of the grid to be simulated.

[0109] Calculate conditional probability: Use indicator kriging (IK) to calculate the class to which the current node belongs. probability .

[0110] Parameter processing: In IK calculation, the parameters of the variation function generated by CNN are directly called to construct the covariance matrix.

[0111] Sampling: The phase type of a point is determined by Monte Carlo sampling from the cumulative distribution function, and it is used as a known point in the calculation of the next node.

[0112] For example, the key parameter intelligent in-property modeling based on neural meshes and variograms in step two is implemented using sequential Gaussian simulations for predicting continuous properties such as porosity and permeability.

[0113] Data preparation: Normal transformation: Converting the original data Z(x) into Y(x) that conforms to a standard normal distribution. N(0,1).

[0114] Synergistic constraints: Seismic inversion attribute volumes are treated as synergistic variables.

[0115] Algorithm formula: at node Conditional mean at and variance Obtained through simple Kriging:

[0116]

[0117] For example, in the automated workflow execution and closed-loop iterative optimization of step three, the system's built-in model quality assessment unit automatically and quantitatively checks the 3D geological model generated in each iteration, quantifying the model's reliability by calculating multiple key indicators. These indicators mainly include: the matching error between the structural layer model and the well point stratification elevation; the consistency between the formation thickness reflected by the structural model and the known geological statistical thickness; the similarity between the histogram of the attribute model's numerical distribution (such as porosity and permeability) and the histogram of the well logging curve; and the matching accuracy between the simulated values ​​of the attribute model at the well location and the actual well logging curve values. If the assessment results show that one or more indicators consistently fail to meet the preset quality standards (such as errors exceeding the threshold or statistical characteristics not matching), the system will automatically trigger a feedback loop mechanism. This mechanism will re-call the parameter generation model, fine-tune the modeling parameters, and generate a new parameter set to drive a new round of modeling calculations. The entire closed-loop process of "modeling → assessment → optimization" will iterate cyclically until all quality indicators of the output 3D geological model meet the predetermined requirements, thereby ensuring the accuracy and geological rationality of the final model.

[0118] For example, the above-mentioned 3D geological modeling method based on neural networks and variograms is applied to Block A for intelligent reservoir 3D geological modeling. Block A is a typical lithofacies oil and gas reservoir controlled by a deltaic sedimentary system in a certain oil and gas basin. The overall structure is a monocline structure with a high northeast and low southwest, good inheritance, and no faults. The modeling area is 12m². 2The total number of wells is 68. This case study utilizes an automated modeling system, integrating multi-source data: well location data, well trajectory data, stratification data, well logging data, and seismic interpretation stratigraphic data. After data standardization, a pre-defined 3D reservoir geological modeling workflow is established via a graphical interface. The workflow sets the model name as "Block A 3D Reservoir Geological Model," the grid step size to 10×10, the vertical dimension to include 4 structural planes, and the longitudinal dimension to divide into 24 layers. An automatic model evaluation strategy and evaluation indicators are set, and the automated modeling workflow is invoked to generate the structural model and attribute model. The workflow first calls the intelligent calculation interface for structural modeling parameters, inputting the seismic interpretation stratigraphic data and well stratification data into a pre-trained CNN model. The CNN automatically predicts the variogram parameters required for Kriging interpolation (main range 1142 meters, secondary range 447 meters, nugget value 0.01, sill 1, azimuth 42 degrees) by analyzing the spatial continuity characteristics of the input data, generating a 3D structural framework model with a total grid count of 6.193 million. Based on the structural model, well logging curves are input into the facies model and attribute model. Following the model's recommended sequential indicator simulation and sequential Gaussian simulation variability functions, a lithofacies model and a porosity-permeability-saturation attribute model are generated. The calculated structural model achieves a 92% agreement rate with well point stratification. Comparison between the generated attribute model and well logging interpretation results shows an average error of less than 3%, and the data distribution error between the model attribute distribution histogram and the well logging curve attribute distribution histogram is less than 5%. The model quality is excellent, and the geological regularities are reasonable. Specifically, as shown below... Figure 2 As shown.

[0119] Therefore, this embodiment provides a three-dimensional geological modeling method based on neural networks and variograms, which has the following advantages compared to existing modeling methods: First, precision, automation, and practicality: While respecting existing geological modeling work standards, the focus is on solving the problem of automatically generating key parameters that rely most heavily on human experience in the workflow. By deeply integrating a neural network-based intelligent calculation method for variograms into each stage of the standard workflow, a key parameter generation mechanism centered on automated processes is constructed, significantly improving modeling efficiency and automation levels, and is easily integrated with existing industrial software platforms.

[0120] Second, the geological rationality of parameter generation: By leveraging the powerful feature extraction and pattern recognition capabilities of CNN deep learning models, complex nonlinear relationships can be learned from multi-source geological data, thereby generating algorithm parameters with clearer geological significance and more in line with the actual underground conditions, improving the prediction accuracy and reliability of the model.

[0121] Third, the transparency and controllability of the process: Due to the pre-defined and interpretable workflow, the entire modeling process is transparent and auditable to the user. Geological experts can easily intervene in the workflow settings, verify the intelligently generated parameters, and make manual adjustments when necessary, realizing intelligent modeling through human-machine collaboration and enhancing the credibility of the results.

[0122] Fourth, the system's adaptive optimization capability: The built-in evaluation unit enables the system to construct a highly automated "modeling-evaluation-optimization" closed loop, giving it significant self-learning and adaptive capabilities. The evaluation unit quantitatively assesses the quality of the 3D geological model generated in each iteration. If the result does not meet the preset standards, the system will automatically trigger parameter optimization instructions, feeding back to the intelligent parameter calculation module to fine-tune key modeling parameters and drive a new round of modeling calculations. This continuous iteration mechanism based on real-time feedback effectively reduces reliance on external manual intervention, ensuring that the system can continuously approach the optimal solution in multiple iterations, thereby significantly improving the accuracy and geological rationality of the final model.

[0123] like Figure 4 As shown, this embodiment also provides a three-dimensional geological modeling system based on neural networks and variograms, including: a workflow construction module, used to construct a structured three-dimensional geological modeling workflow based on the geological background of the work area and the modeling objectives; the three-dimensional geological modeling workflow is used to reflect the complete sequence of steps and dependencies from data input to three-dimensional geological model output; a model training module, used to train a pre-constructed initial network model using pre-processed multi-source geological data to obtain a trained parameter generation model, and to generate key parameters required for each stage of the structured three-dimensional geological modeling workflow based on the parameter generation model; wherein, the initial network model is constructed based on neural networks and variograms; and an execution module, used to call the key parameters and execute the structured three-dimensional geological modeling workflow. During the execution of the structured three-dimensional geological modeling workflow, a quantitative quality assessment is performed on the generated intermediate geological model. If the assessment indicators do not meet the preset standards, the parameter generation model is used to fine-tune the parameters, and the structured three-dimensional geological modeling workflow and assessment process are repeated until a three-dimensional geological model that meets the quality requirements is generated, and finally, the three-dimensional geological model is output.

[0124] The present invention also provides a three-dimensional geological modeling device based on neural networks and variograms, comprising: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of the three-dimensional geological modeling method based on neural networks and variograms.

[0125] The present invention also provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the three-dimensional geological modeling method based on neural networks and variograms.

[0126] When the processor executes the computer program, it implements the steps of the above-mentioned three-dimensional geological modeling based on neural networks and variograms, for example: constructing a structured three-dimensional geological modeling workflow based on the geological background and modeling objectives of the work area; the three-dimensional geological modeling workflow is used to reflect the complete sequence of steps and dependencies from data input to three-dimensional geological model output; training the pre-constructed initial network model with pre-processed multi-source geological data to obtain a trained parameter generation model, and generating key parameters required for each stage of the structured three-dimensional geological modeling workflow based on the parameter generation model; wherein, the initial network model is constructed based on neural networks and variograms; calling the key parameters and executing the structured three-dimensional geological modeling workflow; during the execution of the structured three-dimensional geological modeling workflow, performing a quantitative quality assessment on the generated intermediate geological model; if the assessment indicators do not meet the preset standards, the parameter generation model is used to fine-tune the parameters, and the structured three-dimensional geological modeling workflow and assessment process are repeated until a three-dimensional geological model that meets the quality requirements is generated, and finally the three-dimensional geological model is output.

[0127] For example, the computer program can be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units can be a series of computer program instruction segments capable of performing preset functions, wherein the instruction segments describe the execution process of the computer program in the three-dimensional geological modeling device based on neural networks and variograms. For example, the computer program can be divided into a workflow construction module, a model training module, and an execution module; wherein: the workflow construction module is used to construct a structured 3D geological modeling workflow based on the geological background and modeling objectives of the work area; the 3D geological modeling workflow is used to reflect the complete sequence of steps and dependencies from data input to 3D geological model output; the model training module is used to train a pre-constructed initial network model using pre-processed multi-source geological data to obtain a trained parameter generation model, and to generate key parameters required for each stage of the structured 3D geological modeling workflow based on the parameter generation model; wherein the initial network model is constructed based on a neural network and a variation function; the execution module is used to call the key parameters and execute the structured 3D geological modeling workflow. During the execution of the structured 3D geological modeling workflow, a quantitative quality assessment is performed on the generated intermediate geological model. If the assessment indicators do not meet the preset standards, the parameter generation model is used to fine-tune the parameters, and the structured 3D geological modeling workflow and assessment process are repeated until a 3D geological model that meets the quality requirements is generated, and finally, the 3D geological model is output.

[0128] The 3D geological modeling device based on neural networks and variograms can be a desktop computer, laptop, handheld computer, or cloud server, etc. This device may include, but is not limited to, processors and memory. Those skilled in the art will understand that the above examples of 3D geological modeling devices based on neural networks and variograms do not constitute a limitation on such devices. They may include more components than described above, or combine certain components, or use different components. For example, the 3D geological modeling device based on neural networks and variograms may also include input / output devices, network access devices, buses, etc.

[0129] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor, or any conventional processor. The processor is the control center of the neural network and variogram-based 3D geological modeling system, connecting various parts of the system via various interfaces and lines.

[0130] The memory can be used to store the computer program and / or modules. The processor realizes various functions of the three-dimensional geological modeling device based on neural networks and variograms by running or executing the computer program and / or modules stored in the memory and calling the data stored in the memory.

[0131] The memory may primarily include a program storage area and a data storage area. The program storage area may store the operating system and at least one application program required for a function (such as sound playback, image playback, etc.). The data storage area may store data created based on the use of the mobile phone (such as audio data, phonebook, etc.). Furthermore, the memory may include high-speed random access memory and non-volatile memory, such as hard disks, RAM, plug-in hard disks, smart media cards (SMC), secure digital cards (SD cards), flash cards, at least one disk storage device, flash memory device, or other volatile solid-state storage devices.

[0132] The present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the three-dimensional geological modeling method based on neural networks and variograms.

[0133] If the modules / units of the three-dimensional geological modeling system based on neural networks and variograms are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium.

[0134] Based on this understanding, the present invention can implement all or part of the processes in the above-mentioned three-dimensional geological modeling method based on neural networks and variograms. This can also be accomplished by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium. When executed by a processor, the computer program can implement the steps of the above-mentioned three-dimensional geological modeling method based on neural networks and variograms. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or preset intermediate forms, etc.

[0135] The computer-readable storage medium may include: any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc.

[0136] It should be noted that the content contained in the computer-readable storage medium may be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable storage medium does not include electrical carrier signals and telecommunication signals.

[0137] The above embodiments are merely one of the implementation methods for achieving the technical solution of the present invention. The scope of protection claimed by the present invention is not limited to this embodiment, but also includes any variations, substitutions and other implementation methods that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention.

[0138] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the present invention.

Claims

1. A three-dimensional geological modeling method based on neural networks and variograms, characterized in that, include: Based on the geological background of the work area and the modeling objectives, a structured 3D geological modeling workflow is constructed. The 3D geological modeling workflow is used to reflect the complete sequence of steps and dependencies from data input to 3D geological model output; The pre-constructed initial network model is trained using pre-processed multi-source geological data to obtain a trained parameter generation model. Based on the parameter generation model, key parameters required for each stage of the structured 3D geological modeling workflow are generated. The initial network model is constructed based on a neural network and a variation function. The process involves calling key parameters and executing a structured 3D geological modeling workflow. During the execution of the structured 3D geological modeling workflow, a quantitative quality assessment is performed on the generated intermediate geological model. If the assessment indicators do not meet the preset standards, a parameter generation model is used to fine-tune the parameters. The structured 3D geological modeling workflow and assessment process are repeated until a 3D geological model that meets the quality requirements is generated, and finally, the 3D geological model is output.

2. The three-dimensional geological modeling method based on neural networks and variograms according to claim 1, characterized in that, The step of constructing a structured 3D geological modeling workflow based on the geological background and modeling objectives of the work area includes a workflow parsing process, an execution sequence arrangement process, and a task scheduling and execution process; wherein: The workflow parsing process includes: reading the workflow steps preset by the graphical interface or script, converting the complete process of data loading → surface modeling → phase modeling → attribute modeling → model verification and output into a logical structure that the system can recognize and process, and marking the task nodes and their sequential dependencies; The execution order arrangement process includes: planning the optimal execution route based on task dependencies, and arranging parallel execution for tasks with no dependencies; The task scheduling and execution process includes: automatically starting each modeling stage according to the planned sequence, passing the output results of the previous task to the next task, and monitoring the running, completion, and failure status of each task in real time to achieve automated workflow.

3. The three-dimensional geological modeling method based on neural networks and variograms according to claim 1, characterized in that, In the step of training the pre-constructed initial network model using pre-processed multi-source geological data, the multi-source geological data includes well location data, well trajectory data, well logging curve data, well stratification data, seismic interpretation stratigraphic data, and seismic inversion data volumes. The preprocessing process of the multi-source geological data includes data type standardization, spatial alignment and datum unification, quality control, well logging curve coarsening, standardization, and data fusion; specifically as follows: The fields of the multi-source geological data are supplemented and the units are labeled according to the preset specifications, and the outliers are filled with standard identifiers. All data are unified to the target projection coordinate system of the work area, and the logging depth is converted into absolute vertical depth through well trajectory data; Invalid data and extreme points were removed through integrity checks, outlier filtering using the interquartile range method, and statistical analysis. For continuous attribute data, the arithmetic mean method is used, and for discrete attributes, the mode method is used to map well logging attributes to three-dimensional grid cells. Perform Zscore standardization on various types of data; Standardized data is spliced ​​together along the channel dimension to construct a four-dimensional input tensor, thereby completing the data fusion.

4. The three-dimensional geological modeling method based on neural networks and variograms according to claim 1, characterized in that, The process of training a pre-constructed initial network model using pre-processed multi-source geological data includes: The pre-collected historical work area sample set is preprocessed into multi-source geological data to serve as training data. The output label adopts the optimal parameter combination of the variogram function, which includes the primary, secondary, and vertical range, the primary and secondary azimuth angles and vertical dip angle, the sill value, and the nugget value. A supervised learning approach is adopted, using backpropagation algorithm and gradient descent optimizer, with mean squared error as the loss function, to minimize the difference between the predicted value and the true label, so that the initial network model learns the nonlinear mapping relationship between data features and variogram parameters, and outputs parameters to generate the model; After training, through forward propagation, feature extraction from convolutional layers, introduction of nonlinearity using the ReLU activation function, dimensionality reduction using pooling layers, and mapping using fully connected layers, the predicted values ​​of the variogram parameters are output as key parameters required for each stage of the structured 3D geological modeling workflow.

5. The three-dimensional geological modeling method based on neural networks and variograms according to claim 1, characterized in that, The structured 3D geological modeling workflow employs the Kriging interpolation algorithm in its construction modeling phase. The specific steps are as follows: A spherical variogram model is constructed using the predicted values ​​of the variogram parameters output by the parameter generation model; the predicted values ​​of the variogram parameters include the sill value, nugget value, and range parameter. By constructing a spatial coordinate transformation matrix using azimuth and dip angles, anisotropic space is converted into isotropic space. The weights are obtained by solving the Kriging equations. The stratigraphic data are then smoothed and coarsely interpolated to generate a structural surface, thus completing the structural modeling process.

6. The three-dimensional geological modeling method based on neural networks and variograms according to claim 1, characterized in that, The phase modeling step of the structured 3D geological modeling workflow is executed using a sequential indicator simulation algorithm, and the specific steps are as follows: Define indicator functions to distinguish different lithofacies or sedimentary facies types, and use a parameter generation model to generate range and sedimentary facies zone extension direction parameters for each facies or facies combination; Traverse all grid nodes to be simulated, use the indicator kriging algorithm, call the variogram parameters generated by the parameter generation model to construct the covariance matrix, and calculate the conditional probability that the current node belongs to the corresponding phase type. The phase type of a node is determined from the cumulative distribution function by Monte Carlo sampling, and the node is used as a known point in the calculation of the next node to generate a three-dimensional phase model.

7. The three-dimensional geological modeling method based on neural networks and variograms according to claim 1, characterized in that, In the step of quantitatively evaluating the quality of the generated intermediate geological model, the quality evaluation indicators of the quantitative quality evaluation include the matching error between the structural layer and the well point stratification elevation, the consistency between the formation thickness and the known geological statistical thickness, the similarity between the attribute model and the well logging curve histogram, and the matching accuracy package of the simulated value and the actual well logging value at the well location. If any indicator exceeds the preset threshold, a feedback loop is automatically triggered, the parameter generation model is called again to fine-tune the modeling parameters, a new parameter set is generated to drive a new round of modeling calculations, and the iteration continues until all quality assessment indicators meet the requirements.

8. A three-dimensional geological modeling system based on neural networks and variograms, characterized in that, include: The workflow construction module is used to build a structured 3D geological modeling workflow based on the geological background of the work area and the modeling objectives. The 3D geological modeling workflow is used to reflect the complete sequence of steps and dependencies from data input to 3D geological model output; The model training module is used to train a pre-constructed initial network model using pre-processed multi-source geological data to obtain a trained parameter generation model, and to generate key parameters required for each stage of the structured 3D geological modeling workflow based on the parameter generation model; wherein, the initial network model is constructed based on a neural network and a variation function; The execution module is used to call key parameters and execute the structured 3D geological modeling workflow. During the execution of the structured 3D geological modeling workflow, the generated intermediate geological model is quantitatively evaluated. If the evaluation indicators do not meet the preset standards, the parameter generation model is used to fine-tune the parameters. The structured 3D geological modeling workflow and evaluation process are repeated until a 3D geological model that meets the quality requirements is generated, and finally the 3D geological model is output.

9. A three-dimensional geological modeling device based on neural networks and variograms, characterized in that, include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the steps of the three-dimensional geological modeling method based on neural networks and variograms as described in any one of claims 1-7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it is used to implement the steps of the three-dimensional geological modeling method based on neural networks and variograms as described in any one of claims 1-7.

Images